The Internet has been highly beneficial to health care – radiology included – improving access in remote areas, allowing for faster and better diagnoses, and vastly improving the management and transfer of medical records and images. However, increased connectivity can lead to increased vulnerability to outside interference. Researchers and cybersecurity experts have begun to examine ways to mitigate the risk of cyberattacks in medical imaging before they become a real danger.
A study presented at the annual meeting of the Radiological Society of North America (RSNA), looked at the potential to tamper with mammogram results. The researchers trained a cycle-consistent generative adversarial network (CycleGAN), a type of artificial intelligence application, on 680 mammographic images from 334 patients, to convert images showing cancer to healthy ones and to do the same, in reverse, for the normal control images. They wanted to determine if a CycleGAN could insert or remove cancer-specific features into mammograms in a realistic fashion.
“As doctors, it is our moral duty to first protect our patients from harm,” said Anton S. Becker, M.D., radiology resident at University Hospital Zurich and ETH Zurich, in Switzerland. “For example, as radiologists we are used to protecting patients from unnecessary radiation. When neural networks or other algorithms inevitably find their way into our clinical routine, we will need to learn how to protect our patients from any unwanted side effects of those as well.”
The images were presented to three radiologists, who reviewed the images and indicated whether they thought the images were genuine or modified. None of the radiologists could reliably distinguish between the two. “Neural networks, such as CycleGAN, are not only able to learn what breast cancer looks like,” Becker said, “we have now shown that they can insert these learned characteristics into mammograms of healthy patients or remove cancerous lesions from the image and replace them with normal looking tissue.”
Becker anticipates that this type of attack won’t be feasible for at least five years and said patients shouldn’t be concerned right now. Still, he hopes to draw the attention of the medical community, and hardware and software vendors, so that they may make the necessary adjustments to address this issue while it is still theoretical. He said that artificial intelligence, in general, will greatly enrich radiology, offering faster diagnoses and other advantages. He added that there are positive aspects to these findings as well. “Neural networks can teach us more about the image characteristics of certain cancers, making us better doctors.”
Source: Radiological Society of North America